1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * linux/arch/arm/boot/compressed/head.S
5 * Copyright (C) 1996-2002 Russell King
6 * Copyright (C) 2004 Hyok S. Choi (MPU support)
8 #include <linux/linkage.h>
9 #include <asm/assembler.h>
12 #include "efi-header.S"
14 AR_CLASS( .arch armv7-a )
15 M_CLASS( .arch armv7-m )
20 * Note that these macros must not contain any code which is not
21 * 100% relocatable. Any attempt to do so will result in a crash.
22 * Please select one of the following when turning on debugging.
26 #if defined(CONFIG_DEBUG_ICEDCC)
28 #if defined(CONFIG_CPU_V6) || defined(CONFIG_CPU_V6K) || defined(CONFIG_CPU_V7)
29 .macro loadsp, rb, tmp1, tmp2
31 .macro writeb, ch, rb, tmp
32 mcr p14, 0, \ch, c0, c5, 0
34 #elif defined(CONFIG_CPU_XSCALE)
35 .macro loadsp, rb, tmp1, tmp2
37 .macro writeb, ch, rb, tmp
38 mcr p14, 0, \ch, c8, c0, 0
41 .macro loadsp, rb, tmp1, tmp2
43 .macro writeb, ch, rb, tmp
44 mcr p14, 0, \ch, c1, c0, 0
50 #include CONFIG_DEBUG_LL_INCLUDE
52 .macro writeb, ch, rb, tmp
53 #ifdef CONFIG_DEBUG_UART_FLOW_CONTROL
56 waituarttxrdy \tmp, \rb
61 #if defined(CONFIG_ARCH_SA1100)
62 .macro loadsp, rb, tmp1, tmp2
63 mov \rb, #0x80000000 @ physical base address
64 #ifdef CONFIG_DEBUG_LL_SER3
65 add \rb, \rb, #0x00050000 @ Ser3
67 add \rb, \rb, #0x00010000 @ Ser1
71 .macro loadsp, rb, tmp1, tmp2
72 addruart \rb, \tmp1, \tmp2
90 * Debug kernel copy by printing the memory addresses involved
92 .macro dbgkc, begin, end, cbegin, cend
98 kphex \begin, 8 /* Start of compressed kernel */
102 kphex \end, 8 /* End of compressed kernel */
107 kphex \cbegin, 8 /* Start of kernel copy */
111 kphex \cend, 8 /* End of kernel copy */
117 * Debug print of the final appended DTB location
119 .macro dbgadtb, begin, end
127 kphex \begin, 8 /* Start of appended DTB */
132 kphex \end, 8 /* End of appended DTB */
138 .macro enable_cp15_barriers, reg
139 mrc p15, 0, \reg, c1, c0, 0 @ read SCTLR
140 tst \reg, #(1 << 5) @ CP15BEN bit set?
142 orr \reg, \reg, #(1 << 5) @ CP15 barrier instructions
143 mcr p15, 0, \reg, c1, c0, 0 @ write SCTLR
144 ARM( .inst 0xf57ff06f @ v7+ isb )
150 * The kernel build system appends the size of the
151 * decompressed kernel at the end of the compressed data
152 * in little-endian form.
154 .macro get_inflated_image_size, res:req, tmp1:req, tmp2:req
155 adr \res, .Linflated_image_size_offset
157 add \tmp1, \tmp1, \res @ address of inflated image size
159 ldrb \res, [\tmp1] @ get_unaligned_le32
160 ldrb \tmp2, [\tmp1, #1]
161 orr \res, \res, \tmp2, lsl #8
162 ldrb \tmp2, [\tmp1, #2]
163 ldrb \tmp1, [\tmp1, #3]
164 orr \res, \res, \tmp2, lsl #16
165 orr \res, \res, \tmp1, lsl #24
168 .macro be32tocpu, val, tmp
170 /* convert to little endian */
171 eor \tmp, \val, \val, ror #16
172 bic \tmp, \tmp, #0x00ff0000
173 mov \val, \val, ror #8
174 eor \val, \val, \tmp, lsr #8
178 .section ".start", "ax"
180 * sort out different calling conventions
184 * Always enter in ARM state for CPUs that support the ARM ISA.
185 * As of today (2014) that's exactly the members of the A and R
190 .type start,#function
192 * These 7 nops along with the 1 nop immediately below for
193 * !THUMB2 form 8 nops that make the compressed kernel bootable
194 * on legacy ARM systems that were assuming the kernel in a.out
195 * binary format. The boot loaders on these systems would
196 * jump 32 bytes into the image to skip the a.out header.
197 * with these 8 nops filling exactly 32 bytes, things still
198 * work as expected on these legacy systems. Thumb2 mode keeps
199 * 7 of the nops as it turns out that some boot loaders
200 * were patching the initial instructions of the kernel, i.e
201 * had started to exploit this "patch area".
206 #ifndef CONFIG_THUMB2_KERNEL
209 AR_CLASS( sub pc, pc, #3 ) @ A/R: switch to Thumb2 mode
210 M_CLASS( nop.w ) @ M: already in Thumb2 mode
215 .word _magic_sig @ Magic numbers to help the loader
216 .word _magic_start @ absolute load/run zImage address
217 .word _magic_end @ zImage end address
218 .word 0x04030201 @ endianness flag
219 .word 0x45454545 @ another magic number to indicate
220 .word _magic_table @ additional data table
224 ARM_BE8( setend be ) @ go BE8 if compiled for BE8
225 AR_CLASS( mrs r9, cpsr )
226 #ifdef CONFIG_ARM_VIRT_EXT
227 bl __hyp_stub_install @ get into SVC mode, reversibly
229 mov r7, r1 @ save architecture ID
230 mov r8, r2 @ save atags pointer
232 #ifndef CONFIG_CPU_V7M
234 * Booting from Angel - need to enter SVC mode and disable
235 * FIQs/IRQs (numeric definitions from angel arm.h source).
236 * We only do this if we were in user mode on entry.
238 mrs r2, cpsr @ get current mode
239 tst r2, #3 @ not user?
241 mov r0, #0x17 @ angel_SWIreason_EnterSVC
242 ARM( swi 0x123456 ) @ angel_SWI_ARM
243 THUMB( svc 0xab ) @ angel_SWI_THUMB
245 safe_svcmode_maskall r0
246 msr spsr_cxsf, r9 @ Save the CPU boot mode in
250 * Note that some cache flushing and other stuff may
251 * be needed here - is there an Angel SWI call for this?
255 * some architecture specific code can be inserted
256 * by the linker here, but it should preserve r7, r8, and r9.
261 #ifdef CONFIG_AUTO_ZRELADDR
263 * Find the start of physical memory. As we are executing
264 * without the MMU on, we are in the physical address space.
265 * We just need to get rid of any offset by aligning the
268 * This alignment is a balance between the requirements of
269 * different platforms - we have chosen 128MB to allow
270 * platforms which align the start of their physical memory
271 * to 128MB to use this feature, while allowing the zImage
272 * to be placed within the first 128MB of memory on other
273 * platforms. Increasing the alignment means we place
274 * stricter alignment requirements on the start of physical
275 * memory, but relaxing it means that we break people who
276 * are already placing their zImage in (eg) the top 64MB
280 and r4, r4, #0xf8000000
281 /* Determine final kernel image address. */
282 add r4, r4, #TEXT_OFFSET
288 * Set up a page table only if it won't overwrite ourself.
289 * That means r4 < pc || r4 - 16k page directory > &_end.
290 * Given that r4 > &_end is most unfrequent, we add a rough
291 * additional 1MB of room for a possible appended DTB.
298 orrcc r4, r4, #1 @ remember we skipped cache_on
307 get_inflated_image_size r9, r10, lr
309 #ifndef CONFIG_ZBOOT_ROM
310 /* malloc space is above the relocated stack (64k max) */
311 add r10, sp, #MALLOC_SIZE
314 * With ZBOOT_ROM the bss/stack is non relocatable,
315 * but someone could still run this code from RAM,
316 * in which case our reference is _edata.
321 mov r5, #0 @ init dtb size to 0
322 #ifdef CONFIG_ARM_APPENDED_DTB
324 * r4 = final kernel address (possibly with LSB set)
325 * r5 = appended dtb size (still unknown)
327 * r7 = architecture ID
328 * r8 = atags/device tree pointer
329 * r9 = size of decompressed image
330 * r10 = end of this image, including bss/stack/malloc space if non XIP
333 * if there are device trees (dtb) appended to zImage, advance r10 so that the
334 * dtb data will get relocated along with the kernel if necessary.
339 ldr r1, =0xedfe0dd0 @ sig is 0xd00dfeed big endian
344 bne dtb_check_done @ not found
346 #ifdef CONFIG_ARM_ATAG_DTB_COMPAT
348 * OK... Let's do some funky business here.
349 * If we do have a DTB appended to zImage, and we do have
350 * an ATAG list around, we want the later to be translated
351 * and folded into the former here. No GOT fixup has occurred
352 * yet, but none of the code we're about to call uses any
356 /* Get the initial DTB size */
360 /* 50% DTB growth should be good enough */
361 add r5, r5, r5, lsr #1
362 /* preserve 64-bit alignment */
365 /* clamp to 32KB min and 1MB max */
370 /* temporarily relocate the stack past the DTB work space */
379 * If returned value is 1, there is no ATAG at the location
380 * pointed by r8. Try the typical 0x100 offset from start
381 * of RAM and hope for the best.
384 sub r0, r4, #TEXT_OFFSET
394 mov r8, r6 @ use the appended device tree
397 * Make sure that the DTB doesn't end up in the final
398 * kernel's .bss area. To do so, we adjust the decompressed
399 * kernel size to compensate if that .bss size is larger
400 * than the relocated code.
402 ldr r5, =_kernel_bss_size
403 adr r1, wont_overwrite
408 /* Get the current DTB size */
412 /* preserve 64-bit alignment */
416 /* relocate some pointers past the appended dtb */
424 * Check to see if we will overwrite ourselves.
425 * r4 = final kernel address (possibly with LSB set)
426 * r9 = size of decompressed image
427 * r10 = end of this image, including bss/stack/malloc space if non XIP
429 * r4 - 16k page directory >= r10 -> OK
430 * r4 + image length <= address of wont_overwrite -> OK
431 * Note: the possible LSB in r4 is harmless here.
437 adr r9, wont_overwrite
442 * Relocate ourselves past the end of the decompressed kernel.
444 * r10 = end of the decompressed kernel
445 * Because we always copy ahead, we need to do it from the end and go
446 * backward in case the source and destination overlap.
449 * Bump to the next 256-byte boundary with the size of
450 * the relocation code added. This avoids overwriting
451 * ourself when the offset is small.
453 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
456 /* Get start of code we want to copy and align it down. */
460 /* Relocate the hyp vector base if necessary */
461 #ifdef CONFIG_ARM_VIRT_EXT
463 and r0, r0, #MODE_MASK
468 * Compute the address of the hyp vectors after relocation.
469 * This requires some arithmetic since we cannot directly
470 * reference __hyp_stub_vectors in a PC-relative way.
471 * Call __hyp_set_vectors with the new address so that we
472 * can HVC again after the copy.
475 movw r1, #:lower16:__hyp_stub_vectors - 0b
476 movt r1, #:upper16:__hyp_stub_vectors - 0b
484 sub r9, r6, r5 @ size to copy
485 add r9, r9, #31 @ rounded up to a multiple
486 bic r9, r9, #31 @ ... of 32 bytes
494 * We are about to copy the kernel to a new memory area.
495 * The boundaries of the new memory area can be found in
496 * r10 and r9, whilst r5 and r6 contain the boundaries
497 * of the memory we are going to copy.
498 * Calling dbgkc will help with the printing of this
501 dbgkc r5, r6, r10, r9
504 1: ldmdb r6!, {r0 - r3, r10 - r12, lr}
506 stmdb r9!, {r0 - r3, r10 - r12, lr}
509 /* Preserve offset to relocated code. */
512 mov r0, r9 @ start of relocated zImage
513 add r1, sp, r6 @ end of relocated zImage
522 ldmia r0, {r1, r2, r3, r11, r12}
523 sub r0, r0, r1 @ calculate the delta offset
526 * If delta is zero, we are running at the address we were linked at.
530 * r4 = kernel execution address (possibly with LSB set)
531 * r5 = appended dtb size (0 if not present)
532 * r7 = architecture ID
544 #ifndef CONFIG_ZBOOT_ROM
546 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
547 * we need to fix up pointers into the BSS region.
548 * Note that the stack pointer has already been fixed up.
554 * Relocate all entries in the GOT table.
555 * Bump bss entries to _edata + dtb size
557 1: ldr r1, [r11, #0] @ relocate entries in the GOT
558 add r1, r1, r0 @ This fixes up C references
559 cmp r1, r2 @ if entry >= bss_start &&
560 cmphs r3, r1 @ bss_end > entry
561 addhi r1, r1, r5 @ entry += dtb size
562 str r1, [r11], #4 @ next entry
566 /* bump our bss pointers too */
573 * Relocate entries in the GOT table. We only relocate
574 * the entries that are outside the (relocated) BSS region.
576 1: ldr r1, [r11, #0] @ relocate entries in the GOT
577 cmp r1, r2 @ entry < bss_start ||
578 cmphs r3, r1 @ _end < entry
579 addlo r1, r1, r0 @ table. This fixes up the
580 str r1, [r11], #4 @ C references.
585 not_relocated: mov r0, #0
586 1: str r0, [r2], #4 @ clear bss
594 * Did we skip the cache setup earlier?
595 * That is indicated by the LSB in r4.
603 * The C runtime environment should now be setup sufficiently.
604 * Set up some pointers, and start decompressing.
605 * r4 = kernel execution address
606 * r7 = architecture ID
610 mov r1, sp @ malloc space above stack
611 add r2, sp, #MALLOC_SIZE @ 64k max
615 get_inflated_image_size r1, r2, r3
617 mov r0, r4 @ start of inflated image
618 add r1, r1, r0 @ end of inflated image
622 #ifdef CONFIG_ARM_VIRT_EXT
623 mrs r0, spsr @ Get saved CPU boot mode
624 and r0, r0, #MODE_MASK
625 cmp r0, #HYP_MODE @ if not booted in HYP mode...
626 bne __enter_kernel @ boot kernel directly
628 adr r12, .L__hyp_reentry_vectors_offset
633 __HVC(0) @ otherwise bounce to hyp mode
635 b . @ should never be reached
638 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
646 .word __bss_start @ r2
648 .word _got_start @ r11
653 LC1: .word .L_user_stack_end - LC1 @ sp
654 .word _edata - LC1 @ r6
658 .word _end - restart + 16384 + 1024*1024
660 .Linflated_image_size_offset:
661 .long (input_data_end - 4) - .
663 #ifdef CONFIG_ARCH_RPC
665 params: ldr r0, =0x10000100 @ params_phys for RPC
672 * dcache_line_size - get the minimum D-cache line size from the CTR register
675 .macro dcache_line_size, reg, tmp
676 #ifdef CONFIG_CPU_V7M
677 movw \tmp, #:lower16:BASEADDR_V7M_SCB + V7M_SCB_CTR
678 movt \tmp, #:upper16:BASEADDR_V7M_SCB + V7M_SCB_CTR
681 mrc p15, 0, \tmp, c0, c0, 1 @ read ctr
684 and \tmp, \tmp, #0xf @ cache line size encoding
685 mov \reg, #4 @ bytes per word
686 mov \reg, \reg, lsl \tmp @ actual cache line size
690 * Turn on the cache. We need to setup some page tables so that we
691 * can have both the I and D caches on.
693 * We place the page tables 16k down from the kernel execution address,
694 * and we hope that nothing else is using it. If we're using it, we
698 * r4 = kernel execution address
699 * r7 = architecture number
702 * r0, r1, r2, r3, r9, r10, r12 corrupted
703 * This routine must preserve:
707 cache_on: mov r3, #8 @ cache_on function
711 * Initialize the highest priority protection region, PR7
712 * to cover all 32bit address and cacheable and bufferable.
714 __armv4_mpu_cache_on:
715 mov r0, #0x3f @ 4G, the whole
716 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
717 mcr p15, 0, r0, c6, c7, 1
720 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
721 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
722 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
725 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
726 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
729 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
730 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
731 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
732 mrc p15, 0, r0, c1, c0, 0 @ read control reg
733 @ ...I .... ..D. WC.M
734 orr r0, r0, #0x002d @ .... .... ..1. 11.1
735 orr r0, r0, #0x1000 @ ...1 .... .... ....
737 mcr p15, 0, r0, c1, c0, 0 @ write control reg
740 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
741 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
744 __armv3_mpu_cache_on:
745 mov r0, #0x3f @ 4G, the whole
746 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
749 mcr p15, 0, r0, c2, c0, 0 @ cache on
750 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
753 mcr p15, 0, r0, c5, c0, 0 @ access permission
756 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
758 * ?? ARMv3 MMU does not allow reading the control register,
759 * does this really work on ARMv3 MPU?
761 mrc p15, 0, r0, c1, c0, 0 @ read control reg
762 @ .... .... .... WC.M
763 orr r0, r0, #0x000d @ .... .... .... 11.1
764 /* ?? this overwrites the value constructed above? */
766 mcr p15, 0, r0, c1, c0, 0 @ write control reg
768 /* ?? invalidate for the second time? */
769 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
772 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
778 __setup_mmu: sub r3, r4, #16384 @ Page directory size
779 bic r3, r3, #0xff @ Align the pointer
782 * Initialise the page tables, turning on the cacheable and bufferable
783 * bits for the RAM area only.
787 mov r9, r9, lsl #18 @ start of RAM
788 add r10, r9, #0x10000000 @ a reasonable RAM size
789 mov r1, #0x12 @ XN|U + section mapping
790 orr r1, r1, #3 << 10 @ AP=11
792 1: cmp r1, r9 @ if virt > start of RAM
793 cmphs r10, r1 @ && end of RAM > virt
794 bic r1, r1, #0x1c @ clear XN|U + C + B
795 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
796 orrhs r1, r1, r6 @ set RAM section settings
797 str r1, [r0], #4 @ 1:1 mapping
802 * If ever we are running from Flash, then we surely want the cache
803 * to be enabled also for our execution instance... We map 2MB of it
804 * so there is no map overlap problem for up to 1 MB compressed kernel.
805 * If the execution is in RAM then we would only be duplicating the above.
807 orr r1, r6, #0x04 @ ensure B is set for this
811 orr r1, r1, r2, lsl #20
812 add r0, r3, r2, lsl #2
819 @ Enable unaligned access on v6, to allow better code generation
820 @ for the decompressor C code:
821 __armv6_mmu_cache_on:
822 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
823 bic r0, r0, #2 @ A (no unaligned access fault)
824 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
825 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
826 b __armv4_mmu_cache_on
828 __arm926ejs_mmu_cache_on:
829 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
830 mov r0, #4 @ put dcache in WT mode
831 mcr p15, 7, r0, c15, c0, 0
834 __armv4_mmu_cache_on:
837 mov r6, #CB_BITS | 0x12 @ U
840 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
841 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
842 mrc p15, 0, r0, c1, c0, 0 @ read control reg
843 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
845 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
846 bl __common_mmu_cache_on
848 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
852 __armv7_mmu_cache_on:
853 enable_cp15_barriers r11
856 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
858 movne r6, #CB_BITS | 0x02 @ !XN
861 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
863 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
865 mrc p15, 0, r0, c1, c0, 0 @ read control reg
866 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
867 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
868 orr r0, r0, #0x003c @ write buffer
869 bic r0, r0, #2 @ A (no unaligned access fault)
870 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
871 @ (needed for ARM1176)
873 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
874 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
875 orrne r0, r0, #1 @ MMU enabled
876 movne r1, #0xfffffffd @ domain 0 = client
877 bic r6, r6, #1 << 31 @ 32-bit translation system
878 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
879 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
880 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
881 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
883 mcr p15, 0, r0, c7, c5, 4 @ ISB
884 mcr p15, 0, r0, c1, c0, 0 @ load control register
885 mrc p15, 0, r0, c1, c0, 0 @ and read it back
887 mcr p15, 0, r0, c7, c5, 4 @ ISB
892 mov r6, #CB_BITS | 0x12 @ U
895 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
896 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
897 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
898 mrc p15, 0, r0, c1, c0, 0 @ read control reg
899 orr r0, r0, #0x1000 @ I-cache enable
900 bl __common_mmu_cache_on
902 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
905 __common_mmu_cache_on:
906 #ifndef CONFIG_THUMB2_KERNEL
908 orr r0, r0, #0x000d @ Write buffer, mmu
911 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
912 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
914 .align 5 @ cache line aligned
915 1: mcr p15, 0, r0, c1, c0, 0 @ load control register
916 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
917 sub pc, lr, r0, lsr #32 @ properly flush pipeline
920 #define PROC_ENTRY_SIZE (4*5)
923 * Here follow the relocatable cache support functions for the
924 * various processors. This is a generic hook for locating an
925 * entry and jumping to an instruction at the specified offset
926 * from the start of the block. Please note this is all position
936 call_cache_fn: adr r12, proc_types
937 #ifdef CONFIG_CPU_CP15
938 mrc p15, 0, r9, c0, c0 @ get processor ID
939 #elif defined(CONFIG_CPU_V7M)
941 * On v7-M the processor id is located in the V7M_SCB_CPUID
942 * register, but as cache handling is IMPLEMENTATION DEFINED on
943 * v7-M (if existant at all) we just return early here.
944 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
945 * __armv7_mmu_cache_{on,off,flush}) would be selected which
946 * use cp15 registers that are not implemented on v7-M.
950 ldr r9, =CONFIG_PROCESSOR_ID
952 1: ldr r1, [r12, #0] @ get value
953 ldr r2, [r12, #4] @ get mask
954 eor r1, r1, r9 @ (real ^ match)
956 ARM( addeq pc, r12, r3 ) @ call cache function
957 THUMB( addeq r12, r3 )
958 THUMB( moveq pc, r12 ) @ call cache function
959 add r12, r12, #PROC_ENTRY_SIZE
963 * Table for cache operations. This is basically:
966 * - 'cache on' method instruction
967 * - 'cache off' method instruction
968 * - 'cache flush' method instruction
970 * We match an entry using: ((real_id ^ match) & mask) == 0
972 * Writethrough caches generally only need 'on' and 'off'
973 * methods. Writeback caches _must_ have the flush method
977 .type proc_types,#object
979 .word 0x41000000 @ old ARM ID
988 .word 0x41007000 @ ARM7/710
997 .word 0x41807200 @ ARM720T (writethrough)
999 W(b) __armv4_mmu_cache_on
1000 W(b) __armv4_mmu_cache_off
1004 .word 0x41007400 @ ARM74x
1006 W(b) __armv3_mpu_cache_on
1007 W(b) __armv3_mpu_cache_off
1008 W(b) __armv3_mpu_cache_flush
1010 .word 0x41009400 @ ARM94x
1012 W(b) __armv4_mpu_cache_on
1013 W(b) __armv4_mpu_cache_off
1014 W(b) __armv4_mpu_cache_flush
1016 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
1018 W(b) __arm926ejs_mmu_cache_on
1019 W(b) __armv4_mmu_cache_off
1020 W(b) __armv5tej_mmu_cache_flush
1022 .word 0x00007000 @ ARM7 IDs
1031 @ Everything from here on will be the new ID system.
1033 .word 0x4401a100 @ sa110 / sa1100
1035 W(b) __armv4_mmu_cache_on
1036 W(b) __armv4_mmu_cache_off
1037 W(b) __armv4_mmu_cache_flush
1039 .word 0x6901b110 @ sa1110
1041 W(b) __armv4_mmu_cache_on
1042 W(b) __armv4_mmu_cache_off
1043 W(b) __armv4_mmu_cache_flush
1046 .word 0xffffff00 @ PXA9xx
1047 W(b) __armv4_mmu_cache_on
1048 W(b) __armv4_mmu_cache_off
1049 W(b) __armv4_mmu_cache_flush
1051 .word 0x56158000 @ PXA168
1053 W(b) __armv4_mmu_cache_on
1054 W(b) __armv4_mmu_cache_off
1055 W(b) __armv5tej_mmu_cache_flush
1057 .word 0x56050000 @ Feroceon
1059 W(b) __armv4_mmu_cache_on
1060 W(b) __armv4_mmu_cache_off
1061 W(b) __armv5tej_mmu_cache_flush
1063 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
1064 /* this conflicts with the standard ARMv5TE entry */
1065 .long 0x41009260 @ Old Feroceon
1067 b __armv4_mmu_cache_on
1068 b __armv4_mmu_cache_off
1069 b __armv5tej_mmu_cache_flush
1072 .word 0x66015261 @ FA526
1074 W(b) __fa526_cache_on
1075 W(b) __armv4_mmu_cache_off
1076 W(b) __fa526_cache_flush
1078 @ These match on the architecture ID
1080 .word 0x00020000 @ ARMv4T
1082 W(b) __armv4_mmu_cache_on
1083 W(b) __armv4_mmu_cache_off
1084 W(b) __armv4_mmu_cache_flush
1086 .word 0x00050000 @ ARMv5TE
1088 W(b) __armv4_mmu_cache_on
1089 W(b) __armv4_mmu_cache_off
1090 W(b) __armv4_mmu_cache_flush
1092 .word 0x00060000 @ ARMv5TEJ
1094 W(b) __armv4_mmu_cache_on
1095 W(b) __armv4_mmu_cache_off
1096 W(b) __armv5tej_mmu_cache_flush
1098 .word 0x0007b000 @ ARMv6
1100 W(b) __armv6_mmu_cache_on
1101 W(b) __armv4_mmu_cache_off
1102 W(b) __armv6_mmu_cache_flush
1104 .word 0x000f0000 @ new CPU Id
1106 W(b) __armv7_mmu_cache_on
1107 W(b) __armv7_mmu_cache_off
1108 W(b) __armv7_mmu_cache_flush
1110 .word 0 @ unrecognised type
1119 .size proc_types, . - proc_types
1122 * If you get a "non-constant expression in ".if" statement"
1123 * error from the assembler on this line, check that you have
1124 * not accidentally written a "b" instruction where you should
1125 * have written W(b).
1127 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1128 .error "The size of one or more proc_types entries is wrong."
1132 * Turn off the Cache and MMU. ARMv3 does not support
1133 * reading the control register, but ARMv4 does.
1136 * r0, r1, r2, r3, r9, r12 corrupted
1137 * This routine must preserve:
1141 cache_off: mov r3, #12 @ cache_off function
1144 __armv4_mpu_cache_off:
1145 mrc p15, 0, r0, c1, c0
1147 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1149 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1150 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1151 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1154 __armv3_mpu_cache_off:
1155 mrc p15, 0, r0, c1, c0
1157 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1159 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1162 __armv4_mmu_cache_off:
1164 mrc p15, 0, r0, c1, c0
1166 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1168 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1169 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1173 __armv7_mmu_cache_off:
1174 mrc p15, 0, r0, c1, c0
1180 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1183 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1185 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1186 mcr p15, 0, r0, c7, c10, 4 @ DSB
1187 mcr p15, 0, r0, c7, c5, 4 @ ISB
1191 * Clean and flush the cache to maintain consistency.
1194 * r0 = start address
1195 * r1 = end address (exclusive)
1197 * r1, r2, r3, r9, r10, r11, r12 corrupted
1198 * This routine must preserve:
1207 __armv4_mpu_cache_flush:
1212 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1213 mov r1, #7 << 5 @ 8 segments
1214 1: orr r3, r1, #63 << 26 @ 64 entries
1215 2: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1216 subs r3, r3, #1 << 26
1217 bcs 2b @ entries 63 to 0
1218 subs r1, r1, #1 << 5
1219 bcs 1b @ segments 7 to 0
1222 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1223 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1226 __fa526_cache_flush:
1230 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1231 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1232 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1235 __armv6_mmu_cache_flush:
1238 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1239 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1240 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1241 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1244 __armv7_mmu_cache_flush:
1245 enable_cp15_barriers r10
1248 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1249 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1252 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1255 dcache_line_size r1, r2 @ r1 := dcache min line size
1256 sub r2, r1, #1 @ r2 := line size mask
1257 bic r0, r0, r2 @ round down start to line size
1258 sub r11, r11, #1 @ end address is exclusive
1259 bic r11, r11, r2 @ round down end to line size
1260 0: cmp r0, r11 @ finished?
1262 mcr p15, 0, r0, c7, c14, 1 @ Dcache clean/invalidate by VA
1266 mcr p15, 0, r10, c7, c10, 4 @ DSB
1267 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1268 mcr p15, 0, r10, c7, c10, 4 @ DSB
1269 mcr p15, 0, r10, c7, c5, 4 @ ISB
1272 __armv5tej_mmu_cache_flush:
1275 1: mrc p15, 0, APSR_nzcv, c7, c14, 3 @ test,clean,invalidate D cache
1277 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1278 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1281 __armv4_mmu_cache_flush:
1284 mov r2, #64*1024 @ default: 32K dcache size (*2)
1285 mov r11, #32 @ default: 32 byte line size
1286 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1287 teq r3, r9 @ cache ID register present?
1292 mov r2, r2, lsl r1 @ base dcache size *2
1293 tst r3, #1 << 14 @ test M bit
1294 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1298 mov r11, r11, lsl r3 @ cache line size in bytes
1301 bic r1, r1, #63 @ align to longest cache line
1304 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1305 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1306 THUMB( add r1, r1, r11 )
1310 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1311 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1312 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1315 __armv3_mmu_cache_flush:
1316 __armv3_mpu_cache_flush:
1320 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1324 * Various debugging routines for printing hex characters and
1325 * memory, which again must be relocatable.
1329 .type phexbuf,#object
1331 .size phexbuf, . - phexbuf
1333 @ phex corrupts {r0, r1, r2, r3}
1334 phex: adr r3, phexbuf
1348 @ puts corrupts {r0, r1, r2, r3}
1349 puts: loadsp r3, r2, r1
1350 1: ldrb r2, [r0], #1
1353 2: writeb r2, r3, r1
1363 @ putc corrupts {r0, r1, r2, r3}
1370 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1371 memdump: mov r12, r0
1374 2: mov r0, r11, lsl #2
1382 ldr r0, [r12, r11, lsl #2]
1402 #ifdef CONFIG_ARM_VIRT_EXT
1404 __hyp_reentry_vectors:
1407 #ifdef CONFIG_EFI_STUB
1408 W(b) __enter_kernel_from_hyp @ hvc from HYP
1414 W(b) __enter_kernel @ hyp
1417 #endif /* CONFIG_ARM_VIRT_EXT */
1420 mov r0, #0 @ must be 0
1421 mov r1, r7 @ restore architecture number
1422 mov r2, r8 @ restore atags pointer
1423 ARM( mov pc, r4 ) @ call kernel
1424 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1425 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1429 #ifdef CONFIG_EFI_STUB
1430 __enter_kernel_from_hyp:
1431 mrc p15, 4, r0, c1, c0, 0 @ read HSCTLR
1432 bic r0, r0, #0x5 @ disable MMU and caches
1433 mcr p15, 4, r0, c1, c0, 0 @ write HSCTLR
1437 ENTRY(efi_enter_kernel)
1438 mov r4, r0 @ preserve image base
1439 mov r8, r1 @ preserve DT pointer
1441 ARM( adrl r0, call_cache_fn )
1442 THUMB( adr r0, call_cache_fn )
1443 adr r1, 0f @ clean the region of code we
1444 bl cache_clean_flush @ may run with the MMU off
1446 #ifdef CONFIG_ARM_VIRT_EXT
1448 @ The EFI spec does not support booting on ARM in HYP mode,
1449 @ since it mandates that the MMU and caches are on, with all
1450 @ 32-bit addressable DRAM mapped 1:1 using short descriptors.
1452 @ While the EDK2 reference implementation adheres to this,
1453 @ U-Boot might decide to enter the EFI stub in HYP mode
1454 @ anyway, with the MMU and caches either on or off.
1456 mrs r0, cpsr @ get the current mode
1457 msr spsr_cxsf, r0 @ record boot mode
1458 and r0, r0, #MODE_MASK @ are we running in HYP mode?
1462 mrc p15, 4, r1, c1, c0, 0 @ read HSCTLR
1463 tst r1, #0x1 @ MMU enabled at HYP?
1467 @ When running in HYP mode with the caches on, we're better
1468 @ off just carrying on using the cached 1:1 mapping that the
1469 @ firmware provided. Set up the HYP vectors so HVC instructions
1470 @ issued from HYP mode take us to the correct handler code. We
1471 @ will disable the MMU before jumping to the kernel proper.
1473 adr r0, __hyp_reentry_vectors
1474 mcr p15, 4, r0, c12, c0, 0 @ set HYP vector base (HVBAR)
1479 @ When running in HYP mode with the caches off, we need to drop
1480 @ into SVC mode now, and let the decompressor set up its cached
1481 @ 1:1 mapping as usual.
1483 1: mov r9, r4 @ preserve image base
1484 bl __hyp_stub_install @ install HYP stub vectors
1485 safe_svcmode_maskall r1 @ drop to SVC mode
1486 msr spsr_cxsf, r0 @ record boot mode
1487 orr r4, r9, #1 @ restore image base and set LSB
1491 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
1492 tst r0, #0x1 @ MMU enabled?
1493 orreq r4, r4, #1 @ set LSB if not
1496 mov r0, r8 @ DT start
1497 add r1, r8, r2 @ DT end
1498 bl cache_clean_flush
1500 adr r0, 0f @ switch to our stack
1504 mov r5, #0 @ appended DTB size
1505 mov r7, #0xFFFFFFFF @ machine ID
1507 ENDPROC(efi_enter_kernel)
1508 0: .long .L_user_stack_end - .
1512 .section ".stack", "aw", %nobits
1513 .L_user_stack: .space 4096